SnapRef v2a vs v1b comparison review: pose following enhanced but still has flaws
Start here
lilting channel published a review saying that SnapRef v2a, compared with v1b on M1 Max, shows significantly enhanced reference pose following; in the Base running scene, 4 out of 5 seeds turned to the front, but quantitative metrics are basically on par with v1a, and on Turbo there is still leg clothing drift, and MPS side-by-side reference images still output pure black. v2a node automatically maps 2.9B weights, no need to manually remap keys, and Base generation speed is slightly faster than v1b.
Written by AI from the reports · updated 9 days ago
Report timeline
Follow the reports to see the event from other sides.
- lilting channel (English) · 创作分享SnapRef v2a vs v1b single-character reference comparison review: M1 Max real-world test
SnapRef v2a's comparison test with v1b on M1 Max shows that v2a's reference pose following is significantly enhanced; in the Base running scene, 4 out of 5 seeds turned to the front, but quantitative metrics are basically on par with v1a, and on Turbo there is still leg clothing drift, and MPS side-by-side reference images still output pure black. v2a node automatically maps 2.9B weights, no need to manually remap keys, and Base generation speed is slightly faster than v1b.
Heat trend for this event
Not enough consecutive observations to draw a trend yet.